Search

What is Qwen-Coder? Alibaba’s AI models for writing code

This tutorial explains why general chat models fail on real code, introduces Alibaba's Qwen-Coder family and its sizes, and describes how code tokenizers work, with code examples, architecture diagrams, and step-by-step walkthroughs.

What AI labs report vs what independent tests find · whatmodel.sh

A curated list of AI model benchmarks and model-card references, juxtaposing vendor-reported scores with independent test results across numerous models and tasks.

Contextual self model · jbsalles/Selmem@b9fd382 · GitHub

Code defines a contextual self model with roles, evidence types, policy defaults, memory structures, and (de)serialization logic, plus validation rules for scores and context.

Code agents: panorama and selection criteria | SFEIR

Translated title. Original: Agents de code : panorama et critères de choix | SFEIR

The document profiles 33 code-agent tools, highlighting that model choice is only one factor; license terms, execution behavior, jurisdiction, and context access matter more. It notes frequent model and harness changes across vendors and provides examples of model mappings and licensing regimes.

Why Agents Are Soon Only Going to Have 1 Tool

Code Mode collapses multiple tools into one exec(code) tool, enabling a model to write a script that chains, parallelizes, and verifies actions across systems in one pass, driven by code-trained models and servers exposing input/output schemas.

Codecast by Armature — Video Code Walkthroughs for Claude Code & Codex

Codecast provides narrated, visual explanations of code by connecting a coding agent with Claude Code or Codex to explain repositories, PRs, and plans. It combines code, diagrams, and narration, with a setup flow that involves account sign-in, prompt setup, and controlled sharing of code context.

OpenCode 2.1: How the coding agent works according to modular construction principle | heise online

Translated title. Original: OpenCode 2.1: So funktioniert der Coding-Agent nach Baukastenprinzip | heise online

OpenCode 2.1 is an open, model-agnostic coding-agent platform managed by a central Harness service. It supports multiple models (including local ones), integrates via various interfaces, and uses JavaScript-enabled tool calls; upgrading requires plugin updates and some features were removed.

Command Code Docs | The best coding agent for open models

A concise list of feature updates and capabilities for Command Code, covering sessions, permissions, collaboration, testing, prompts, design modes, and codebase context.

China’s AI model “Qwen4” finally arrives: Preview version now available through Alibaba’s AI service (Page 1 of 2) - ITmedia AI+

Translated title. Original: 中国AI「Qwen4」ついに登場　プレビュー版を提供開始、AlibabaのAIサービスで（1/2 ページ） - ITmedia AI＋

Alibaba Cloud has announced Qwen4 and says it improves coding, real-world understanding, and independent task completion. A Qwen4-Max preview is available through its Token Plan and several Alibaba services.

han/han-coding at main · testdouble/han · GitHub

Han coding provides a structured, plugin-backed set of coding skills and workflows (TDD, refactoring, API design, code reviews) powered by specialist agents to handle code changes and quality assurance across a project.